OpenAI quietly gave US government early access to its powerful GPT-5.5 model for ‘national security testing’

OpenAI had provided early access to its GPT-5.5 AI model to the US government for national security testing and evaluations. The revelation was made by Chris Lehane, OpenAI’s vice president of global affairs in a post on LinkedIn.

Lehane also noted that the company is working with US government's Center for AI Standards and Innovation (CAISI) for testing its upcoming models like GPT-5.5 Cyber, an AI model that is designed for cyber defenders.

Just a few days ago, OpenAI CEO Sam Altman had confirmed that the GPT-5.5 Cyber model would be released to “critical cyber defenders” in a few days.

“We will work with the entire ecosystem and the government to figure out trusted access for cyber; we want to rapidly help secure companies and infrastructure,” Altman wrote in a post on X

Meanwhile, Lehane in his new LinkedIn post also confirmed that OpenAI is partnering with White House and the broader US government for deploying AI responsibly.

“we’re partnering with the White House and the broader Administration on a responsible deployment strategy, including a playbook to help get these capabilities into the hands of federal, state, and local governments, allies, and critical infrastructure operators” Lehane wrote

“This reflects a broader AI resiliency need we highlighted in Industrial Policy for the Intelligence Age: building trusted institutions, testing capacity, and international networks that can support both innovation and security. CAISI’s growing network of frontier lab partners is an important step in that direction” he added

The news about OpenAI giving early access to its top of the line AI models comes on the same day when the company announced that it is updating the default model on ChatGPT to GPT-5.5 Instant.

OpenAI's partnerships with US government:

OpenAI has been increasing its involvement with the US government ever since it launched "OpenAI for Government," last year. The company was at the receiving end of public backlash earlier in the year when it signed a contract with the US Department of War (DoW) for deploying its AI models in classified work.

The backlash became even more intense because OpenAI's news of partnership came just hours after it was revealed that Anthropic's deal with Pentagon fell through due to diagreements about safer use of AI.

The model also comes with ‘agentic’ capabilities, meaning it can plan tasks, use tools, check its own work, and continue working through problems with minimal user intervention.

OpenAI also claimed that GPT-5.5 outperformed Anthropic’s Claude Opus 4.7 and Google’s Gemini 3.1 Pro on several benchmarks. For instance, on the Terminal-Bench 2.0 benchmark, which tests complex command-line and coding workflows, GPT-5.5 scored 82.7 per cent, ahead of Claude Opus 4.7’s 69.4 per cent and Gemini 3.1 Pro’s 68.5 per cent

Similarly on FrontierMath Tier 4, an advanced mathematics benchmark, GPT-5.5 scored 35.4 per cent, outperforming Claude Opus 4.7’s 22.9 per cent and Gemini 3.1 Pro’s 16.7 per cent.